Google expands partnership with Intel for AI chips

In a significant move, Google has announced an expanded collaboration with Intel, committing to utilize multiple generations of Intel's central processing units (CPUs) in its artificial intelligence data centers. This partnership builds on a long-standing relationship that dates back nearly 30 years to Google's initial server rack projects. The latest advancement involves Intel's Xeon 6 CPUs, which are set to handle AI training and inference tasks. This shift could enhance Intel's standing in the AI marketplace, which has recently been dominated by Nvidia. "Their Xeon roadmap gives us confidence that we can continue to meet the growing performance and efficiency demands of our workloads," stated Amin Vahdat, Google's Chief Technologist for AI Infrastructure, in a Thursday announcement. While the financial specifics of the agreement were not disclosed, the partnership arrives at a pivotal moment in the AI landscape, where CPU capabilities are becoming increasingly crucial. Dion Harris, Nvidia's head of AI infrastructure, noted earlier in the year that CPUs are starting to emerge as a bottleneck, as the demands of agentic workloads extend beyond traditional graphics processing units. Intel CEO Lip-Bu Tan emphasized the importance of balanced systems in AI development, stating, "Scaling AI requires more than accelerators — it requires balanced systems." This remark highlights the necessity for a comprehensive approach to AI infrastructure. Amid ongoing challenges to keep up with technological advancements, Intel recently sold a 10% stake to the U.S. government, which highlighted the company’s capacity to produce advanced chips domestically. Additionally, Nvidia's announcement of a $5 billion investment in Intel has contributed to a significant rise in Intel's stock, which has nearly tripled in value over the past year. At the core of Intel's advancements is its latest Xeon processor, manufactured using cutting-edge 18A technology at its newly opened fabrication plant in Arizona. Despite heavy investments in its foundry operations, Intel’s own processors remain a primary customer at this facility. In a recent LinkedIn post, Tan revealed that Elon Musk has engaged Intel to design and manufacture custom chips for SpaceX, xAI, and Tesla as part of Musk's ambitious Terafab project in Texas, although details of the financial arrangement and timeline were not provided. Moreover, Google and Intel reiterated their ongoing collaboration on a novel chip known as the infrastructure processing unit (IPU), which has been in development since 2022. According to Intel, this programmable accelerator is designed to offload networking, storage, and security tasks from host CPUs, thereby optimizing the performance of traditional data centers. For over a decade, Google has been developing its own custom AI accelerator, the tensor processing unit (TPU). Additionally, in 2024, Google ventured into creating its own custom CPU named Axion, opting for an Arm-based design rather than sticking with Intel's x86 architecture.